Math 341 Lecture #17

§3.4 Perfect Sets and Connected Sets.

The Cantor set C has another topological property that will prove useful in showing that
C is uncountable.
Definition 3.4.1. A set P ⊂ R is perfect if it is closed and contains no isolated points.
A finite subset of R is closed but it is not perfect.
Closed intervals [c, d] with −∞ < c < d < ∞, are perfect.
What about the Cantor set?
Theorem 3.4.2. The Cantor set C is perfect.
Proof. Each Cn is a finite union of closed intervals, and so is closed.
Then C = ∩Cn is a closed set.
Now we will show that each x ∈ C is not isolated by constructing a sequence (xn ) in C
6 x for all n ∈ N and xn → x.
with xn =
The closed set C1 is the union of two closed intervals I11 and I12 each of length 1/3.
The point x is in one of these two closed intervals, call it I1 .
The intersection C2 ∩ I1 consists of two closed intervals, one of which contains x.
Pick x1 to be an endpoint of the other closed interval, and so x1 6= x.
Because the endpoints of the closed intervals that make C2 are in C, we have that x1 ∈ C.
Because x and x1 are both in I1 and the length of I1 is 1/3, we have |x1 − x| ≤ 1/3.
The closed set C2 is the union of four closed intervals I2j , j = 1, 2, 3, 4, each of length
1/9.
The point x is in one of these four closed intervals, call it I2 .
The intersection C3 ∩ I2 consists of two closed intervals, one of which contains x.
Pick x2 to be an endpoint of the other closed interval, and so x2 6= x.
Because x2 is an endpoint of one of the closed intervals in C3 , we have that x2 ∈ C.
Because x2 and x both belong to I2 which is of length 1/9, we have |x2 − x| ≤ 1/9.
Continuing in this way, we construct a sequence (xn ) in C with xn 6= x for all n ∈ N,
and |xn − x| ≤ 1/3n .
Thus we have shown that x is a limit point of C, and therefore x is not isolated.
As x was an arbitrary point of C, we have that C is perfect. 
In this proof we used the endpoints of the closed intervals in Cn to form a sequence (xn )
that converged to the given point x in the Cantor set.
Each endpoint is rational of the form m/3n for 0 ≤ m ≤ 3n , but this does not mean that
the limit x of (xn ) is rational.
In fact “most” of the sequences formed from the rational endpoints converge to irrational
numbers, and this account for the uncountable nature of the Cantor set.
Theorem 3.4.3. A nonempty perfect set is uncountable.
Proof. A nonempty perfect set P cannot be finite, because in a nonempty finite set each
point is isolated.
So a nonempty perfect set is infinite.
Suppose, to the contrary, that P is countable.
Using a bijection f : N → P we can enumerate the elements of P as

P = {x1 , x2 , x3 , . . . }.

The point here is that every element of P appears in this enumerated list.
Let I1 be a closed interval that contains x1 in its interior (i.e., x1 is not an endpoint of
I1 ).
As P is perfect, the element x1 is not isolated, so there is some other y2 ∈ P such that
y2 is also in the interior of I1 .
Choose a closed interval I2 centered on y2 so that I2 ⊆ I1 and x1 6∈ I2 .
Since y2 ∈ I2 and y2 ∈ P we have I2 ∩ P 6= ∅.
The element y2 ∈ P is not isolated, so there is a y3 ∈ P that is in the interior of I2 .
We may choose y3 6= x2 , for if y3 = x2 then there will be another choice of y3 ∈ P in the
interior of I2 because y3 is not isolated.
Now choose a closed interval I3 ⊆ I2 centered on y3 for which x2 6∈ I3 .
Since y3 ∈ I3 and y3 ∈ P we have I3 ∩ P 6= ∅.
Carrying out this construction inductively results in a sequence of closed intervals In
satisfying

(i) In+1 ⊆ In ,

(ii) xn 6∈ In+1 , and

(iii) In ∩ P 6= ∅.

For each n ∈ N, the set Kn = In ∩ P is compact because In is bounded and In ∩ P closed.

By Theorem 3.3.5 we have that
∞
\
Kn 6= ∅.
n=1

On the other hand, since Kn ⊆ P and xn 6∈ In+1 for all n ∈ N, we have that
∞
\
Kn = ∅.
n=1

This contradiction implies that P is uncountable. 

Now we turn out attention to another topological notion for subsets of R.
Definition 3.4.4. Two nonempty sets A, B ⊆ R are separated if A ∩ B = ∅ and
A ∩ B = ∅.
A set E ⊆ R is disconnected if it can be written as E = A ∪ B for separated sets A and
B.
A set that is not disconnected is called a connected set.
Example 3.4.5. (i) The set A = (1, 2) and B = (2, 5) are separated because
A ∩ B = [1, 2] ∩ (2, 5) = ∅, A ∩ B = (1, 2) ∩ [2, 5] = ∅.

The set E = A ∪ B is disconnected because it is the union of the separated sets A and
B.
On the other hand, the sets C = (1, 2] and D = (2, 5) are not separated because C ∩ D =
{2}.
The set C ∪ D is the interval (1, 5) which is connected (although we have not shown this).
(ii) The set of rational numbers is disconnected.
To see this we set √ √
A = Q ∩ (−∞, 2), B = Q ∩ ( 2, ∞).

We certainly have Q = A ∪ B.
√
The Order Limit Theorem implies that any limit point of A will be in (−∞, 2], which
is disjoint from B.
Similarly, A ∩ B 6= ∅, and so A and B are separated sets.
We conclude that Q is disconnected.
By carefully working through the logical negations of the quantifiers in the definition of
disconnected, we arrive at a positive characterization of connectedness.
Theorem 3.4.6. A set E ⊂ R is connected if and only if, for all nonempty disjoint sets
A and B satisfying E = A ∪ B, there always exists a convergent sequence (xn ) with all
xn contained in one of A or B, and x = lim xn contained in the other.
This notion of connectedness is more relevant in higher dimensions, for in dimension 1,
a subset E ⊆ R is connected precisely when E is an interval.
Theorem 3.4.7. A set E ⊆ R is connected if and only if whenever a < c < b with
a, b ∈ E, it follows for that c ∈ E too.
Proof. Suppose that E is connected, let a, b ∈ E, and pick a < c < b.
Suppose c 6∈ E, and set

A = (−∞, c) ∩ E, B = (c, ∞) ∩ E.

Because a ∈ A and b ∈ B, both A and B are nonempty; and E = A ∪ B.

Since any limit point of l of A satisfies l ≤ c by the Order Limit Theorem, we have that
A ∩ B = ∅.
Similarly, we have A ∩ B = ∅.
Thus A and B are separated set, and so E = A ∪ B is disconnected, a contradiction.
Hence, c ∈ E.
Now suppose whenever a < c < b with a, b ∈ E we have that c ∈ E too.
We will use Theorem 3.4.6 to show that E is connected.
To this end we write E = A ∪ B for nonempty disjoint sets A and B.
Pick a0 ∈ A and b0 ∈ B, and WLOG suppose that a0 < b0 .
Since every c ∈ (a0 , b0 ) must be in E, we have that I0 = [a0 , b0 ] ⊆ E.
Bisect I0 into two equal halves.
The midpoint of I0 is either in A or B.
If the midpoint of I0 is in A, take I1 = [a1 , b1 ] to be the right half where a1 is the midpoint
of I0 and b1 = b0 ∈ B.
If the midpoint of I0 is in B, take I1 = [a1 , b1 ] to be the left half where b1 is the midpoint
of I0 and a1 = a0 ∈ A.
Continuing this process yields a sequence of nested intervals In = [an , bn ] where an ∈ A
and bn ∈ B, and whose lengths bn − an go to 0 as n → ∞.
By the Nested Interval Property, there exists
∞
\
x∈ In .
n=0

The sequence (an ) of left endpoints belongs to A and converges to x, and the sequence
(bn ) of right endpoints belongs to B and converges to x as well.
[Note: (an ) and (bn ) are equivalent Cauchy sequences.]
Since x ∈ I0 and I0 ⊆ E, we have that x ∈ E = A ∪ B, which means that x ∈ A or
x ∈ B.
So there is a limit point of one of A or B that belongs to the other, and by Theorem
3.4.6, the set E is connected.